Watch the game unfold, play by play.
Every game of the 2025 season plus the 2026 season as it unfolds, replayed the way a gamecast should: the drive drawn on the field, the score as it stood at that snap, and the win-chance line riding along underneath. The current season's games land here after each game day — the play-by-play feed posts once games finish, so a 2026 game appears the morning after it's played, never mid-game. These are full game replays built from the real play-by-play — not a live feed. When the live scores feed connects, this same field lights up in real time. Win chance comes from the Lab's model (v1.1.0) — a model, not a certainty; when it says 70%, history says that side wins about 70 times in 100. The 2026 games are priced by that model as out-of-sample: it trained on 2019–2023 and was graded on 2025, and it has never been retrained on this season.
Loading the seasons…
Win chance — home team
The Lab's model (v1.1.0) pricing the game at every snap: score, clock, down, field position, timeouts, and the pregame spread. A model, not a certainty. Tap the line to jump to that play.
How this is built: play-by-play is the nflverse / nflfastR open data (the same pull the Lab trains on), every snap in order. Win chance is computed play by play with the Lab's shipped model v1.1.0 — the same coefficients the What-If Toolbar runs in your browser — and graded on the 2025 season: it beat the public benchmark on 40,610 plays it had never trained on. The 2026 games ride on those same coefficients as out-of-sample games — the model trained on 2019–2023 and was graded on 2025 — and new ones are added as the data posts after each game day. Scores shown on each play are the score the moment after that play. Each season's tie game (Green Bay at Dallas in 2025, Week 4; New York Giants at Washington in 2026, Week 5) isn't here, because the model doesn't price ties.